Kriek, M.T.
Citation
Kriek, M. T. (2007, September 26). The many phases of massive galaxies : a near-infrared spectroscopic study of galaxies in the early universe. Retrieved from
https://hdl.handle.net/1887/12353
Version: Corrected Publisher’s Version
License: Licence agreement concerning inclusion of doctoral thesis in the Institutional Repository of the University of Leiden
Downloaded from: https://hdl.handle.net/1887/12353
Note: To cite this publication please use the final published version (if applicable).
Abstract:
Using the Gemini Near-Infraed Spectrograph (GNIRS), we have completed a near- infrared (NIR) spectroscopic survey for K-bright galaxies at z∼2.3, selected from the multi-wavelength survey by Yale-Chile (MUSYC). We successfully derive spec- troscopic redshifts from emission lines and continuum shapes for all 36 observed galaxies. The continuum redshifts are driven by the Balmer/4000 ˚A break, and have an uncertainty in∆z/(1+z) of<0.019. We use this unique sample to deter- mine, for the first time, how accurately redshifts and other properties of K-selected high redshift galaxies can be determined from broadband photometric data alone.
We find that the photometric redshifts of the galaxies in our sample have a sys- tematic error of 0.08 and a random error of 0.13 in ∆z/(1+z). We show that the systematic error can be reduced by using optimal templates and deep photometry, but that random uncertainties at the level of∼0.05 cannot be eliminated. Turning to stellar population parameters, we find that the spectra lead to significantly im- proved constraints. For most quantities this improvement is about equally driven by the higher spectral resolution and by the much reduced redshift uncertainty.
We find that properties such as the age, AV, current star formation rate, and the star formation history are generally very poorly constrained with broadband data alone. Interestingly stellar masses and mass-to-light ratios are among the most sta- ble parameters, and we show that the spectroscopy supports our previous finding that red galaxies dominate the high mass end of the galaxy population at z=2−3.
Mariska Kriek, Pieter G. van Dokkum, Marijn Franx, Garth D. Illingworth, Danilo Marchesini, Ryan Quadri, Gregory Rudnick, Edward N. Taylor, Natascha M. F ¨orster Schreiber, Eric Gawiser, Ivo Labb´e, Paulina Lira & Stijn Wuyts The Astrophysical Journal, submitted
65
5.1 Introduction
T
HE EXTENSIVE USE of photometric redshifts has greatly enhanced our knowledge of the z =2−3 universe. While color criteria, such as the Lyman break tech- nique (Steidel et al. 1996a,b), distant red galaxy selection (DRGs, Franx et al. 2003;van Dokkum et al. 2003) and BzK selection (Daddi et al. 2004) provide an easy identi- fication of high-redshift galaxies, photometric redshifts allow the study of apparently unbiased samples. Our current understanding of the evolution of the mass-density (e.g., Rudnick et al. 2001, 2003, 2006; Dickinson et al. 2003; Drory et al. 2005) and the luminosity function (e.g., Dahlen et al. 2005; Saracco et al. 2006; Marchesini et al. 2007), the nature of massive high-redshift galaxies (e.g., F ¨orster Schreiber et al. 2004; Labb´e et al. 2005; van Dokkum 2006; Papovich et al. 2006), and galaxy clustering (e.g., Daddi et al. 2003; Quadri et al. 2007a; Foucaud et al. 2007) essentially all rely on photometric redshifts.
Ideally, all these studies would have been based on spectroscopic redshifts. How- ever, obtaining spectroscopic redshifts for the required samples is hampered by several obstacles. Due to their faintness, obtaining redshifts of high-redshift galaxies require long integrations on 8-10m class telescopes. The largest samples of spectroscopically confirmed high redshift galaxies number in the 1000s (e.g., Steidel et al. 2003), several orders of magnitude short of the largest photometric samples. Furthermore, and more fundamentally, current spectroscopic samples are strongly biased towards blue, star forming galaxies which are bright in the observer’s optical (Lyman break galaxies, or LBGs). It has become clear that only∼20% of massive galaxies at z =2−3 are blue LBGs, and the typical massive galaxy at this redshift range is red in the rest-frame optical and faint in the rest-frame ultra-violet (UV) (van Dokkum 2006).
Obtaining spectroscopic redshifts for typical massive galaxies is possible, but it re- quires deep spectroscopy in the near-infrared (NIR). NIR observations are complicated by the combination of the high sky brightness, numerous bright and variable night sky lines and strong atmospheric absorption bands, and the limited size of current and planned detectors in NIR spectrographs. Obtaining continuum detections for hun- dreds or thousands of galaxies with K ∼21 (the typical brightness of galaxies with M>1011M⊙ at z∼2.5) will not be feasible in the foreseeable future. Until the next generation of space missions and>20 m ground-based telescopes we remain largely dependent on photometric redshifts for studies of large and faint galaxy samples be- yond z>1.5.
Our provisional dependency on broadband photometric studies requires a more accurate calibration and understanding of the involved systematics. The current spec- troscopic samples used for calibration of photometric high-redshift studies are based primarily on optical spectroscopy. As these samples are biased towards un-obscured star-forming galaxies, their systematics may not be representative for the total sample of massive galaxies. Photometric properties of red, massive galaxies at high redshift are poorly calibrated, and since red galaxies dominate the high mass end at 2<z<3, systematics may have large effects on the final results.
NIR spectroscopy on a substantial, unbiased sample of massive, high-redshift galax- ies is needed to test our photometric studies, and obtain insights regarding possible
In this chapter we present our full survey of K-selected galaxies at z ∼2.3, con- ducted with GNIRS on Gemini-South between September 2004 and March 2007. In total we integrated more than∼80 hours, divided over 6 observing runs, on a sample of 36 galaxies. In previous chapters based on preliminary results of this survey we dis- cussed the stellar populations (Chapter 3, Kriek et al. 2006b) and the origin of the line emission (Chapter 4, Kriek et al. 2007). In this chapter we will give an overview of the total survey (§ 5.2), compare photometric and spectroscopic redshifts (§ 5.3) and stel- lar populations properties (§ 5.4), and discuss the implications for photometric studies (§ 5.4).
Throughout the chapter we assume aΛCDM cosmology withΩm=0.3, ΩΛ=0.7, and H0 =70 km s−1 Mpc−1. The broadband magnitudes are given in the Vega-based photometric system unless stated otherwise. Furthermore, we will measure the scatter and the offset between various properties using the normalized biweight mean abso- lute deviation and the biweight mean, respectively (Beers, Flynn & Gebhardt 1990). As biweight statistics are less sensitive towards outliers than the normal mean, and more efficient than the median, they are most appropriate for the small sample sizes in this work.
5.2 Data
5.2.1 Sample Selection
The galaxies studied in this work are selected from the multi-wavelength survey by Yale-Chile (MUSYC, Gawiser et al. 2006; Quadri et al. 2007b). This survey consist of optical imaging (UBVRIz) of four 30′ ×30′ fields, shallow NIR imaging (JHK) over the same area, and deeper NIR imaging over four 10′ ×10′ fields. The depth of the deep and wide NIR photometry is K ∼21 and K ∼20 (5σ) respectively. The spectro- scopic follow-up presented in this chapter is selected from the deep fields HDF-South, 1030, and 1256 (Quadri et al. 2007b), and the shallow extended Chandra Deep Field South (ECDFS, E.N. Taylor et al. 2007, in preparation). One galaxy is selected from the Great Observatories Origins Deep Survey (GOODS; Giavalisco et al. 2004). The optical-to-NIR photometry that we used as part of this work is described by S. Wuyts et al. (2007, in preparation).
We selected galaxies with 2.0<∼zphot<
∼2.7 (see § 5.3.2) and K<∼19.7. This redshift interval is chosen as the bright emission lines Hβ, [OIII], Hα, and [NII] fall in the H and K atmospheric windows. A few galaxies had 2.7<∼zphot<
∼3.0 at the time of selection,
Table 5.1 —Sample and Observations
Observation Exptime Old
RA DEC Ks R J−K dates min IDa
1030-32 10 30 41.5 5 19 55 19.68 26.83 2.62 2006/12/19 230 -
2007/03/12
1030-101 10 30 10.1 5 20 11 18.97 25.63 2.15 2006/02/23 120 -
1030-301 10 30 50.8 5 20 49 18.82 25.49 2.81 2006/01/20 90 -
1030-609 10 30 49.6 5 21 50 19.68 24.23 2.02 2006/02/24 130 -
1030-807 10 30 20.0 5 22 33 19.72 24.77 2.40 2006/02/23 120 -
1030-1531 10 30 38.9 5 24 52 19.38 22.92 2.23 2006/02/25 80 -
1030-1611 10 30 48.4 5 25 03 19.58 25.55 2.37 2006/02/24 120 -
1030-1813 10 30 51.2 5 25 36 19.01 24.97 2.93 2006/01/20 80 -
1030-1839 10 30 45.4 5 30 07 19.61 24.20 2.35 2006/12/16 80 -
1030-2026 10 30 22.7 5 28 26 19.48 25.22 2.93 2006/02/22 120 -
1030-2329 10 30 16.2 5 27 32 19.72 25.24 2.47 2006/02/25 120 -
1030-2559 10 30 40.1 5 26 34 19.62 25.89 2.52 2006/02/22 110 -
1030-2728b 10 30 18.4 5 26 05 19.52 25.09 2.69 2006/01/21 120 -
1030-2927 10 30 43.3 5 29 34 19.48 24.52 2.23 2006/12/18 230 -
2007/03/13
1256-0 12 54 59.6 1 11 30 19.26 24.98 2.26 2005/05/19+27+30 305 151 2006/02/24
1256-142 12 55 02.7 1 07 32 19.45 25.99 2.54 2006/02/23 120 465
1256-519 12 55 08.4 1 06 14 18.99 25.51 2.55 2006/02/25 80 -
1256-1207 12 55 19.7 1 12 46 19.25 25.34 2.04 2006/02/25 80 -
1256-1967 12 55 25.8 1 03 25 18.71 23.55 2.02 2005/05/18 240 2889 2006/01/18
HDFS1-259 22 33 11.2 -60 40 47 19.42 24.00 2.11 2006/12/17-18 140 - HDFS1-1849 22 33 37.9 -60 33 15 19.30 25.18 2.51 2004/09/06 115 - HDFS2-509 22 31 23.1 -60 39 08 18.57 23.20 2.50 2005/05/16+19 235 - HDFS2-1099 22 32 03.2 -60 36 13 19.26 24.94 2.55 2006/12/19 120 - HDFS2-2046 22 32 30.8 -60 32 44 19.38 24.24 2.21 2005/05/20 125 - ECDFS-4454 3 32 11.5 -27 55 23 19.24 24.28 2.89 2006/01/18 100 3662 ECDFS-4511 3 32 43.2 -27 55 15 18.77 23.36 2.62 2006/01/21 190 3694
2006/02/25
ECDFS-4713 3 31 52.5 -27 54 48 18.68 22.94 2.13 2006/02/22 60 3896 ECDFS-5856 3 32 13.3 -27 52 26 19.42 25.42 3.21 2006/01/19 120 4937 ECDFS-6842 3 31 51.3 -27 50 56 19.09 24.47 2.47 2006/12/19 210 -
2007/03/11+14
ECDFS-6956 3 32 02.5 -27 50 46 19.16 23.40 2.43 2006/01/20 150 5754 2006/02/24
ECDFS-9822 3 31 33.9 -27 46 03 19.14 24.13 3.04 2006/12/17 120 - ECDFS-11490 3 32 45.0 -27 43 09 19.25 24.01 2.49 2006/01/20+21 190 9510 ECDFS-12514 3 31 39.5 -27 41 20 19.11 22.79 1.72 2006/02/23 90 10525 ECDFS-13532 3 31 54.8 -27 39 23 19.52 24.98 3.51 2006/12/18 160 - ECDFS-16671 3 31 58.9 -27 35 16 18.99 22.08 1.74 2006/12/16 60 - CDFS-6202 3 32 31.5 -27 46 23 19.04 23.62 2.28 2004/09/02+03 90 6036
aID numbers in Chapters 3 and 4 (Kriek et al. 2006b, 2007)
bThe spectroscopic redshift of this galaxy has first been confirmed using K-band spectroscopy with NIRSPEC on Keck, in 2005 January.
Figure 5.1 —Comparison between the photometric properties of the GNIRS sample at 2<zphot<3 and a mass-limited sample (M>1011M⊙) at 2<zphot<
3. The probabilities (P) that the GNIRS sample and the full mass-selected sample are drawn from the same distribution, as derived using a Mann- Whitney (MW) and a Kolmorov-Smirnov (KS) test, are given in the panels. Additionally, we divide the mass-selected sample into its K-bright (K < 19.7) and K-faint (K>19.7) members. The GNIRS sam- ple may be somewhat less representative for a K- bright sample, as the redshift distribution is differ- ent.
a preliminary stage at the time of se- lection may complicate some of the in- terpretation in this work. For this rea- son and the fact that the ECDFS is the only field with shallow NIR photome- try, the analysis in this chapter will also focus on the subsample excluding the ECDFS.
In total we obtained usable NIR spectra for a sample of 36 galaxies. For
∼4 additional galaxies we obtained empty spectra due to mis-alignment or extremely bad weather conditions.
It is important to establish whether our sample is representative for the galaxy population at z ∼2.5. In Fig- ure 5.1 we compare the distributions of J−K, R−K, rest-frame U−V color and zphot of our spectroscopic sample
with a photometric mass-limited sample (>1011M⊙). For the latter we use the deep MUSYC fields, as the wide NIR data are not deep enough to extract a mass-limited sample (van Dokkum 2006). According to a Mann-Whitney and a Kolmorov-Smirnov test, the photometric properties J−K, R−K, rest-frame U−V color (as derived from the photometry) and zphot of the GNIRS sample are representative for a photometric mass-limited sample at 2 <z <3 (see probabilities in panels of Fig. 5.1). The Mann- Whitney test assesses whether the two sample populations are consistent with the same mean of distribution, while the K-S test examines whether the two samples could have been drawn from the same parent distribution. For both tests the probability should be greater than the 0.05 significance level.
The galaxies targeted with GNIRS are all bright in K (<19.7). In order to examine if bright galaxies are a biased sub-sample of the total mass-limited sample, we split the sample in bright and faint members. Figure 5.1 shows that the bright and faint members have the same distribution of rest-frame U-V and observed R−K colors.
The main difference between the bright and faint members is the redshift distribution:
almost all K-bright galaxies have zphot<2.3. This also causes their bluer J−K colors:
Figure 5.2 —Comparison of J−K and R−K col- ors as a function of zphot between a photometric mass-limited (M>1011M⊙) sample and our spec- troscopic sample. The gray diamonds and dots represent all massive galaxies in the deep MUSYC fields (SDSS1030, 1256 and HDF-S) with K<19.7 and K>19.7 respectively. The black symbols repre- sent the 36 galaxies of the GNIRS sample, selected from both the MUSYC deep (diamonds) and wide (ECDFS, crosses) surveys.
in contrast to the R-band, the J-band does not fall entirely bluewards of the optical break for z<2.3. The redshift dependence of J−K is clearly visible in Figure 5.2. However, except for the difference in redshift and presumably stellar mass, we see no hints that the bright and faint members of a mass- limited sample at 2< z <3 have dif- ferent stellar populations. Thus, al- though the spectroscopic sample may have similar stellar population proper- ties as K-bright galaxies, it is less repre- sentative for a K-bright sample at 2<
z < 3, as the median redshift and its corresponding distribution is substan- tially different.
5.2.2 NIR spectra
We observed the full sample of 36 galaxies with GNIRS in cross-dispersed mode, in combination with the 32 lines mm−1 grating and the 0“.675 slit.
This configuration resulted in a spec- tral resolution of R∼1000. The galax- ies were observed during six observ- ing runs in 2004 September (program GS-2004B-Q-38), 2005 May (program GS-2005A-Q-20), 2006 January (program GS- 2005B-C-12) and 2006 February (program GS-2006A-C-6), 2006 December (program GS-2006B-C-5) and 2007 March (program GS-2007A-C-9). During the first two runs most time was lost due to bad weather, and only a handful of galaxies was observed under mediocre weather condition (seeing∼1′′). The weather was excellent through- out the full 3rd and 4th run, and we reached a median seeing of∼0“.5. The conditions were slightly worse during the last two runs, with a median seeing of∼0“.7, and some time was lost due to clouds.
We observed the galaxies following an ABA’B’ on-source dither pattern. In this way we can use the average of the previous and following exposures as sky frame, which cancels sky variation and reduces the noise in the final frame. All targets were acquired by blind offsets from nearby stars. The individual exposures are 5 minutes for the galaxies observed during the first two runs, and 10 minutes for the remaining runs1. The total integration times for all galaxies are listed in Table 5.1. Before and
1An instrument upgrade after the second run improved the throughput and thus the quality of the data, and eliminated “radiation events” caused by radioactive coatings. This allowed longer exposures
Figure 5.4 —Comparison of photometric NIR colors, and those derived from the spectra, for galaxies in the deep (solid) and wide (open) MUSYC fields. Both colors are not corrected for flux contributions by emission lines. The fraction of galaxies for which the photometric and spectroscopic colors are con- sistent within 1σis given in the top left corner. As these factors are<0.68 the errors may be slightly underestimated.
Figure 5.3 —The differences between the low res- olution spectra of two observing sequences for the same galaxy (left: 1030-301, right: 1256-0) are used to estimate a systematic uncertainty on the spectra.
In order to improve the consistency between the dif- ferent observing sequences, we increase the original uncertainty per bin (black error bars) by a systematic error of 10% (gray error bars) of the average flux in the binned spectrum. The fractions of bins that are consistent within 1σfor the original and increased uncertainties are given in the panels in black and gray respectively.
after every observing sequence we ob- serve an AV0 star, for the purpose of correcting for telluric absorption. The final spectra of the two stars were com- bined to match the target’s airmass.
A detailed description of the re- duction procedure of the GNIRS cross- dispersed spectra is given in Chapter 2 (Kriek et al. 2006a). In summary, we subtract the sky, mask cosmic rays and bad pixels, straighten the spec- tra, combine the individual exposures, stitch the orders and finally correct for the response function. 1D spectra are extracted by summing all adjacent lines (along the spatial direction) with a mean flux greater than 0.25 times the flux in the central row, using op- timal weighting. We also constructed
“low resolution” binned spectra from the 2D spectra for each galaxy follow- ing the method as described in Chap- ter 2 (Kriek et al. 2006a). Each bin con- tains 80 “good” pixels (i.e., wavelength regions with high atmospheric trans- mission and low sky emission), corre- sponding to 400 ˚A per bin. Sky, trans- mission, and noise spectra were con-
Figure 5.5 —See caption on page 74.
Figure 5.5 —See caption on page 74.
Figure 5.5 — GNIRS spectra (black squares) and MUSYC broadband photometry (open circles) for all 36 galaxies, sorted for their total K-band magnitude starting with the brightest galaxy (see Table 5.1).
Emission line fluxes are removed from the binned spectra, using the best-fit models to the lines. The photometry is not corrected for emission line contamination. The best-fit stellar population models to the spectra and photometry are shown in gray. For emission-line galaxies the redshift was fixed at zline during fitting, while for the remaining galaxies z was a free parameter. For galaxies with∆z/(1+z)>0.1 we also show the best fit to just the photometry and the corresponding zphot.
structed for each galaxy as well.
We assess the uncertainties on the low resolution spectra by splitting the data in two sequences for several objects and comparing the results. We find that the errors as derived from the photon noise underestimate the true uncertainty for most galaxies. In Figure 5.3 we show two examples. In order to obtain a better consistency between the observing sequences, we increase the uncertainties for all bins by quadratically adding 10% of the average flux in the spectrum.
We use the broadband NIR photometry to perform the absolute flux calibration.
For each galaxy we integrate the spectrum over the same J, H and K filter curves as the photometry. We determine one scaling factor per galaxy, from the difference between the spectroscopic and photometric NIR magnitudes, and use this factor to scale the NIR spectrum. We note that at this stage both the spectra and the photometry may contain flux contributions by emission lines. We extend our wavelength coverage by attaching the optical photometry to the scaled spectra. For the emission-line galaxies we subsequently remove the line fluxes from the affected bins, using the best-fit to the emission lines as derived in Chapter 4 (Kriek et al. 2007).
As a quality check we compare the photometric NIR colors J−H, J−K and H−K to those derived from the spectra. The direct comparison for the individual galaxies is presented in Figure 5.4. The fraction of galaxies for which the spectroscopic and photometric colors are consistent within 1σare listed in the panels. For all colors these
We suspect that the surface brightness of the object plays an important role, as bright objects with low quality spectra and typical integration times, such as ECDFS-4511 and ECDFS-9822, are extended even in the MUSYC imaging data, which have a image quality of∼1′′.
5.3 Spectroscopic versus Photometric Redshifts
5.3.1 Spectroscopic Redshifts and Galaxy Properties
Figure 5.6 — In this figure we illustrate the accu- racy of continuum redshifts, by deriving zcontfor the 19 emission-line galaxies in the sample. The scat- ter and the systematic offset in∆z/(1+z) are listed in the figure. The continuum redshifts are ∼4− 7 times more accurate than photometric redshifts, and show no significant systematic offset. Galaxies without emission lines generally have larger breaks, so their zcontmay even be more accurate.
For 19 of the galaxies in the sam- ple we detected one or more emis- sion lines, and thus for these galax- ies we could determine exact spectro- scopic redshifts. The remaining galax- ies may have no or very faint emis- sion lines, or the lines are expected in atmospheric wavelength regions with low transmission or strong sky emis- sion. Fortunately, we can derive fairly precise redshifts from the continuum emission alone. This is due to the pres- ence of the Balmer/4000 ˚A break in the NIR spectra (Chapter 2, Kriek et al.
2006a). For none of the galaxies ab- sorption lines are detected.
We fit the low resolution binned spectra together with the optical pho- tometry by Bruzual & Charlot (2003) stellar population models. We allow a grid of 24 different ages (not allow- ing the galaxy to be older than the age of the universe), and 31 different exponentially declining star formation histories (SFHs) with the characteristic timescale (τ) varying between 10 Myr
and 10 Gyr. We leave redshift as a free parameter for the galaxies without emission
lines. Furthermore, we adopt the Calzetti et al. (2000) reddening law and allow 41 val- ues for AV between 0 en 4 mag. We compute the χ2 surface as function of all stellar population parameters. For all grid points we assume the Salpeter (1955) IMF, and so- lar metallicity (based on measurements by van Dokkum et al. 2004). A Chabrier (2003) or a Kroupa (2001) IMF yield stellar masses and SFRs which are a factor of∼2 lower.
The mass differences when using the stellar population library by Maraston (2005) are discussed in Kannappan & Gawiser (2007) and Wuyts et al. (2007a).
Figure 5.7 —In this diagram we examine the cause of errors in continuum redshifts. As the modeling is mainly driven by the optical break, we expect and indeed find less accurate continuum redshifts for galaxies for which a large part of the break falls between atmospheric windows or outside the spec- trum (panel a, shaded regions). These galaxies are in- dicated by gray symbols in all panels. In panels b and c we show that for galaxies with bluer SEDs, and thus weaker optical breaks, the continuum red- shifts are less accurate. There is no clear correlation with the S/N of the spectrum in panel d.
We derive 1σ confidence intervals on the redshifts and stellar population properties using 200 Monte Carlo sim- ulations. We vary all bins of the low- resolution spectra according to their uncertainties, and fit the simulated spectra using the same procedure as described above. Next, we determine the contour in the original χ2 surface that encompasses 68% of the Monte Carlo simulations (see Papovich et al.
2003; Kriek et al. 2006a). The 1σ con- fidence intervals for all properties are the minimum and maximum values that are allowed within thisχ2contour.
For the emission-line galaxies, we removed the emission line fluxes from the spectra before fitting. This is differ- ent to our previous method presented in Chapter 2 (Kriek et al. 2006a) in which we mask the bins that are con- taminated by emission lines. The dif- ference in modeling results (although consistent within the errors) compared to Chapters 3 and 4 (Kriek et al. 2006b, 2007) are due to this improvement and catalog updates. All spectroscopic red- shifts and corresponding stellar popu- lation properties are listed in Table 5.2.
In order to test the accuracy of the continuum redshifts, we also fit the emission-line galaxies with redshift as a free parameter. In Figure 5.6 we compare the emission line redshifts with the continuum redshifts. We find a scatter of∆z/(1+z)=0.019 and no significant systematic offset. In Figure 5.7 we examine causes of the errors. First, as the modeling is driven by the optical break, we expect this method to be less accurate if the break falls between atmospheric windows. In the Figure 5.7a we indeed find that galaxies for which the break falls outside the spectrum or in between the J and H band have less accurate zcont.
Furthermore, we expect the continuum redshifts to be more accurate for galaxies
